Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
1da177e4 LT |
3 | * Copyright (C) 1991, 1992 Linus Torvalds |
4 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
5 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
6 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
7 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
8 | * - July2000 | |
1da177e4 LT |
9 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
10 | */ | |
11 | ||
12 | /* | |
13 | * This handles all read/write requests to block devices | |
14 | */ | |
1da177e4 LT |
15 | #include <linux/kernel.h> |
16 | #include <linux/module.h> | |
17 | #include <linux/backing-dev.h> | |
18 | #include <linux/bio.h> | |
19 | #include <linux/blkdev.h> | |
320ae51f | 20 | #include <linux/blk-mq.h> |
1da177e4 LT |
21 | #include <linux/highmem.h> |
22 | #include <linux/mm.h> | |
23 | #include <linux/kernel_stat.h> | |
24 | #include <linux/string.h> | |
25 | #include <linux/init.h> | |
1da177e4 LT |
26 | #include <linux/completion.h> |
27 | #include <linux/slab.h> | |
28 | #include <linux/swap.h> | |
29 | #include <linux/writeback.h> | |
faccbd4b | 30 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 31 | #include <linux/fault-inject.h> |
73c10101 | 32 | #include <linux/list_sort.h> |
e3c78ca5 | 33 | #include <linux/delay.h> |
aaf7c680 | 34 | #include <linux/ratelimit.h> |
6c954667 | 35 | #include <linux/pm_runtime.h> |
eea8f41c | 36 | #include <linux/blk-cgroup.h> |
54d4e6ab | 37 | #include <linux/t10-pi.h> |
18fbda91 | 38 | #include <linux/debugfs.h> |
30abb3a6 | 39 | #include <linux/bpf.h> |
b8e24a93 | 40 | #include <linux/psi.h> |
55782138 LZ |
41 | |
42 | #define CREATE_TRACE_POINTS | |
43 | #include <trace/events/block.h> | |
1da177e4 | 44 | |
8324aa91 | 45 | #include "blk.h" |
43a5e4e2 | 46 | #include "blk-mq.h" |
bd166ef1 | 47 | #include "blk-mq-sched.h" |
bca6b067 | 48 | #include "blk-pm.h" |
c1c80384 | 49 | #include "blk-rq-qos.h" |
8324aa91 | 50 | |
18fbda91 OS |
51 | #ifdef CONFIG_DEBUG_FS |
52 | struct dentry *blk_debugfs_root; | |
53 | #endif | |
54 | ||
d07335e5 | 55 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 56 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 57 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 58 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 59 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
0bfc2455 | 60 | |
a73f730d TH |
61 | DEFINE_IDA(blk_queue_ida); |
62 | ||
1da177e4 LT |
63 | /* |
64 | * For queue allocation | |
65 | */ | |
6728cb0e | 66 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 67 | |
1da177e4 LT |
68 | /* |
69 | * Controlling structure to kblockd | |
70 | */ | |
ff856bad | 71 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 72 | |
8814ce8a BVA |
73 | /** |
74 | * blk_queue_flag_set - atomically set a queue flag | |
75 | * @flag: flag to be set | |
76 | * @q: request queue | |
77 | */ | |
78 | void blk_queue_flag_set(unsigned int flag, struct request_queue *q) | |
79 | { | |
57d74df9 | 80 | set_bit(flag, &q->queue_flags); |
8814ce8a BVA |
81 | } |
82 | EXPORT_SYMBOL(blk_queue_flag_set); | |
83 | ||
84 | /** | |
85 | * blk_queue_flag_clear - atomically clear a queue flag | |
86 | * @flag: flag to be cleared | |
87 | * @q: request queue | |
88 | */ | |
89 | void blk_queue_flag_clear(unsigned int flag, struct request_queue *q) | |
90 | { | |
57d74df9 | 91 | clear_bit(flag, &q->queue_flags); |
8814ce8a BVA |
92 | } |
93 | EXPORT_SYMBOL(blk_queue_flag_clear); | |
94 | ||
95 | /** | |
96 | * blk_queue_flag_test_and_set - atomically test and set a queue flag | |
97 | * @flag: flag to be set | |
98 | * @q: request queue | |
99 | * | |
100 | * Returns the previous value of @flag - 0 if the flag was not set and 1 if | |
101 | * the flag was already set. | |
102 | */ | |
103 | bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) | |
104 | { | |
57d74df9 | 105 | return test_and_set_bit(flag, &q->queue_flags); |
8814ce8a BVA |
106 | } |
107 | EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); | |
108 | ||
2a4aa30c | 109 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 110 | { |
1afb20f3 FT |
111 | memset(rq, 0, sizeof(*rq)); |
112 | ||
1da177e4 | 113 | INIT_LIST_HEAD(&rq->queuelist); |
63a71386 | 114 | rq->q = q; |
a2dec7b3 | 115 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
116 | INIT_HLIST_NODE(&rq->hash); |
117 | RB_CLEAR_NODE(&rq->rb_node); | |
63a71386 | 118 | rq->tag = -1; |
bd166ef1 | 119 | rq->internal_tag = -1; |
522a7775 | 120 | rq->start_time_ns = ktime_get_ns(); |
09e099d4 | 121 | rq->part = NULL; |
b554db14 | 122 | refcount_set(&rq->ref, 1); |
1da177e4 | 123 | } |
2a4aa30c | 124 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 125 | |
e47bc4ed CK |
126 | #define REQ_OP_NAME(name) [REQ_OP_##name] = #name |
127 | static const char *const blk_op_name[] = { | |
128 | REQ_OP_NAME(READ), | |
129 | REQ_OP_NAME(WRITE), | |
130 | REQ_OP_NAME(FLUSH), | |
131 | REQ_OP_NAME(DISCARD), | |
132 | REQ_OP_NAME(SECURE_ERASE), | |
133 | REQ_OP_NAME(ZONE_RESET), | |
6e33dbf2 | 134 | REQ_OP_NAME(ZONE_RESET_ALL), |
6c1b1da5 AJ |
135 | REQ_OP_NAME(ZONE_OPEN), |
136 | REQ_OP_NAME(ZONE_CLOSE), | |
137 | REQ_OP_NAME(ZONE_FINISH), | |
e47bc4ed CK |
138 | REQ_OP_NAME(WRITE_SAME), |
139 | REQ_OP_NAME(WRITE_ZEROES), | |
140 | REQ_OP_NAME(SCSI_IN), | |
141 | REQ_OP_NAME(SCSI_OUT), | |
142 | REQ_OP_NAME(DRV_IN), | |
143 | REQ_OP_NAME(DRV_OUT), | |
144 | }; | |
145 | #undef REQ_OP_NAME | |
146 | ||
147 | /** | |
148 | * blk_op_str - Return string XXX in the REQ_OP_XXX. | |
149 | * @op: REQ_OP_XXX. | |
150 | * | |
151 | * Description: Centralize block layer function to convert REQ_OP_XXX into | |
152 | * string format. Useful in the debugging and tracing bio or request. For | |
153 | * invalid REQ_OP_XXX it returns string "UNKNOWN". | |
154 | */ | |
155 | inline const char *blk_op_str(unsigned int op) | |
156 | { | |
157 | const char *op_str = "UNKNOWN"; | |
158 | ||
159 | if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) | |
160 | op_str = blk_op_name[op]; | |
161 | ||
162 | return op_str; | |
163 | } | |
164 | EXPORT_SYMBOL_GPL(blk_op_str); | |
165 | ||
2a842aca CH |
166 | static const struct { |
167 | int errno; | |
168 | const char *name; | |
169 | } blk_errors[] = { | |
170 | [BLK_STS_OK] = { 0, "" }, | |
171 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
172 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
173 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
174 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
175 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
176 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
177 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
178 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
179 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
86ff7c2a | 180 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
03a07c92 | 181 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
2a842aca | 182 | |
4e4cbee9 CH |
183 | /* device mapper special case, should not leak out: */ |
184 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
185 | ||
2a842aca CH |
186 | /* everything else not covered above: */ |
187 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
188 | }; | |
189 | ||
190 | blk_status_t errno_to_blk_status(int errno) | |
191 | { | |
192 | int i; | |
193 | ||
194 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
195 | if (blk_errors[i].errno == errno) | |
196 | return (__force blk_status_t)i; | |
197 | } | |
198 | ||
199 | return BLK_STS_IOERR; | |
200 | } | |
201 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
202 | ||
203 | int blk_status_to_errno(blk_status_t status) | |
204 | { | |
205 | int idx = (__force int)status; | |
206 | ||
34bd9c1c | 207 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
208 | return -EIO; |
209 | return blk_errors[idx].errno; | |
210 | } | |
211 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
212 | ||
178cc590 CH |
213 | static void print_req_error(struct request *req, blk_status_t status, |
214 | const char *caller) | |
2a842aca CH |
215 | { |
216 | int idx = (__force int)status; | |
217 | ||
34bd9c1c | 218 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
219 | return; |
220 | ||
178cc590 | 221 | printk_ratelimited(KERN_ERR |
b0e5168a CK |
222 | "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x " |
223 | "phys_seg %u prio class %u\n", | |
178cc590 | 224 | caller, blk_errors[idx].name, |
b0e5168a CK |
225 | req->rq_disk ? req->rq_disk->disk_name : "?", |
226 | blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)), | |
227 | req->cmd_flags & ~REQ_OP_MASK, | |
228 | req->nr_phys_segments, | |
229 | IOPRIO_PRIO_CLASS(req->ioprio)); | |
2a842aca CH |
230 | } |
231 | ||
5bb23a68 | 232 | static void req_bio_endio(struct request *rq, struct bio *bio, |
2a842aca | 233 | unsigned int nbytes, blk_status_t error) |
1da177e4 | 234 | { |
78d8e58a | 235 | if (error) |
4e4cbee9 | 236 | bio->bi_status = error; |
797e7dbb | 237 | |
e8064021 | 238 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
b7c44ed9 | 239 | bio_set_flag(bio, BIO_QUIET); |
08bafc03 | 240 | |
f79ea416 | 241 | bio_advance(bio, nbytes); |
7ba1ba12 | 242 | |
143a87f4 | 243 | /* don't actually finish bio if it's part of flush sequence */ |
e8064021 | 244 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
4246a0b6 | 245 | bio_endio(bio); |
1da177e4 | 246 | } |
1da177e4 | 247 | |
1da177e4 LT |
248 | void blk_dump_rq_flags(struct request *rq, char *msg) |
249 | { | |
aebf526b CH |
250 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
251 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
5953316d | 252 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 253 | |
83096ebf TH |
254 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
255 | (unsigned long long)blk_rq_pos(rq), | |
256 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
257 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
258 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 259 | } |
1da177e4 LT |
260 | EXPORT_SYMBOL(blk_dump_rq_flags); |
261 | ||
1da177e4 LT |
262 | /** |
263 | * blk_sync_queue - cancel any pending callbacks on a queue | |
264 | * @q: the queue | |
265 | * | |
266 | * Description: | |
267 | * The block layer may perform asynchronous callback activity | |
268 | * on a queue, such as calling the unplug function after a timeout. | |
269 | * A block device may call blk_sync_queue to ensure that any | |
270 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 271 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
272 | * that its ->make_request_fn will not re-add plugging prior to calling |
273 | * this function. | |
274 | * | |
da527770 | 275 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 276 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 277 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 278 | * |
1da177e4 LT |
279 | */ |
280 | void blk_sync_queue(struct request_queue *q) | |
281 | { | |
70ed28b9 | 282 | del_timer_sync(&q->timeout); |
4e9b6f20 | 283 | cancel_work_sync(&q->timeout_work); |
1da177e4 LT |
284 | } |
285 | EXPORT_SYMBOL(blk_sync_queue); | |
286 | ||
c9254f2d | 287 | /** |
cd84a62e | 288 | * blk_set_pm_only - increment pm_only counter |
c9254f2d | 289 | * @q: request queue pointer |
c9254f2d | 290 | */ |
cd84a62e | 291 | void blk_set_pm_only(struct request_queue *q) |
c9254f2d | 292 | { |
cd84a62e | 293 | atomic_inc(&q->pm_only); |
c9254f2d | 294 | } |
cd84a62e | 295 | EXPORT_SYMBOL_GPL(blk_set_pm_only); |
c9254f2d | 296 | |
cd84a62e | 297 | void blk_clear_pm_only(struct request_queue *q) |
c9254f2d | 298 | { |
cd84a62e BVA |
299 | int pm_only; |
300 | ||
301 | pm_only = atomic_dec_return(&q->pm_only); | |
302 | WARN_ON_ONCE(pm_only < 0); | |
303 | if (pm_only == 0) | |
304 | wake_up_all(&q->mq_freeze_wq); | |
c9254f2d | 305 | } |
cd84a62e | 306 | EXPORT_SYMBOL_GPL(blk_clear_pm_only); |
c9254f2d | 307 | |
165125e1 | 308 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
309 | { |
310 | kobject_put(&q->kobj); | |
311 | } | |
d86e0e83 | 312 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 313 | |
aed3ea94 JA |
314 | void blk_set_queue_dying(struct request_queue *q) |
315 | { | |
8814ce8a | 316 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); |
aed3ea94 | 317 | |
d3cfb2a0 ML |
318 | /* |
319 | * When queue DYING flag is set, we need to block new req | |
320 | * entering queue, so we call blk_freeze_queue_start() to | |
321 | * prevent I/O from crossing blk_queue_enter(). | |
322 | */ | |
323 | blk_freeze_queue_start(q); | |
324 | ||
344e9ffc | 325 | if (queue_is_mq(q)) |
aed3ea94 | 326 | blk_mq_wake_waiters(q); |
055f6e18 ML |
327 | |
328 | /* Make blk_queue_enter() reexamine the DYING flag. */ | |
329 | wake_up_all(&q->mq_freeze_wq); | |
aed3ea94 JA |
330 | } |
331 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
332 | ||
c9a929dd TH |
333 | /** |
334 | * blk_cleanup_queue - shutdown a request queue | |
335 | * @q: request queue to shutdown | |
336 | * | |
c246e80d BVA |
337 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
338 | * put it. All future requests will be failed immediately with -ENODEV. | |
c94a96ac | 339 | */ |
6728cb0e | 340 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 341 | { |
bae85c15 BVA |
342 | WARN_ON_ONCE(blk_queue_registered(q)); |
343 | ||
3f3299d5 | 344 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
aed3ea94 | 345 | blk_set_queue_dying(q); |
6ecf23af | 346 | |
57d74df9 CH |
347 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
348 | blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
349 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); | |
c9a929dd | 350 | |
c246e80d BVA |
351 | /* |
352 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
67ed8b73 BVA |
353 | * prevent that blk_mq_run_hw_queues() accesses the hardware queues |
354 | * after draining finished. | |
c246e80d | 355 | */ |
3ef28e83 | 356 | blk_freeze_queue(q); |
c57cdf7a ML |
357 | |
358 | rq_qos_exit(q); | |
359 | ||
57d74df9 | 360 | blk_queue_flag_set(QUEUE_FLAG_DEAD, q); |
c9a929dd | 361 | |
5a48fc14 DW |
362 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
363 | blk_flush_integrity(); | |
364 | ||
c9a929dd | 365 | /* @q won't process any more request, flush async actions */ |
dc3b17cc | 366 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
c9a929dd TH |
367 | blk_sync_queue(q); |
368 | ||
344e9ffc | 369 | if (queue_is_mq(q)) |
c7e2d94b | 370 | blk_mq_exit_queue(q); |
a1ce35fa | 371 | |
c3e22192 ML |
372 | /* |
373 | * In theory, request pool of sched_tags belongs to request queue. | |
374 | * However, the current implementation requires tag_set for freeing | |
375 | * requests, so free the pool now. | |
376 | * | |
377 | * Queue has become frozen, there can't be any in-queue requests, so | |
378 | * it is safe to free requests now. | |
379 | */ | |
380 | mutex_lock(&q->sysfs_lock); | |
381 | if (q->elevator) | |
382 | blk_mq_sched_free_requests(q); | |
383 | mutex_unlock(&q->sysfs_lock); | |
384 | ||
3ef28e83 | 385 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 386 | |
c9a929dd | 387 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
388 | blk_put_queue(q); |
389 | } | |
1da177e4 LT |
390 | EXPORT_SYMBOL(blk_cleanup_queue); |
391 | ||
165125e1 | 392 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 393 | { |
6d469642 | 394 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
1946089a CL |
395 | } |
396 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 397 | |
3a0a5299 BVA |
398 | /** |
399 | * blk_queue_enter() - try to increase q->q_usage_counter | |
400 | * @q: request queue pointer | |
401 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT | |
402 | */ | |
9a95e4ef | 403 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
3ef28e83 | 404 | { |
cd84a62e | 405 | const bool pm = flags & BLK_MQ_REQ_PREEMPT; |
3a0a5299 | 406 | |
3ef28e83 | 407 | while (true) { |
3a0a5299 | 408 | bool success = false; |
3ef28e83 | 409 | |
818e0fa2 | 410 | rcu_read_lock(); |
3a0a5299 BVA |
411 | if (percpu_ref_tryget_live(&q->q_usage_counter)) { |
412 | /* | |
cd84a62e BVA |
413 | * The code that increments the pm_only counter is |
414 | * responsible for ensuring that that counter is | |
415 | * globally visible before the queue is unfrozen. | |
3a0a5299 | 416 | */ |
cd84a62e | 417 | if (pm || !blk_queue_pm_only(q)) { |
3a0a5299 BVA |
418 | success = true; |
419 | } else { | |
420 | percpu_ref_put(&q->q_usage_counter); | |
421 | } | |
422 | } | |
818e0fa2 | 423 | rcu_read_unlock(); |
3a0a5299 BVA |
424 | |
425 | if (success) | |
3ef28e83 DW |
426 | return 0; |
427 | ||
3a0a5299 | 428 | if (flags & BLK_MQ_REQ_NOWAIT) |
3ef28e83 DW |
429 | return -EBUSY; |
430 | ||
5ed61d3f | 431 | /* |
1671d522 | 432 | * read pair of barrier in blk_freeze_queue_start(), |
5ed61d3f | 433 | * we need to order reading __PERCPU_REF_DEAD flag of |
d3cfb2a0 ML |
434 | * .q_usage_counter and reading .mq_freeze_depth or |
435 | * queue dying flag, otherwise the following wait may | |
436 | * never return if the two reads are reordered. | |
5ed61d3f ML |
437 | */ |
438 | smp_rmb(); | |
439 | ||
1dc3039b | 440 | wait_event(q->mq_freeze_wq, |
7996a8b5 | 441 | (!q->mq_freeze_depth && |
0d25bd07 BVA |
442 | (pm || (blk_pm_request_resume(q), |
443 | !blk_queue_pm_only(q)))) || | |
1dc3039b | 444 | blk_queue_dying(q)); |
3ef28e83 DW |
445 | if (blk_queue_dying(q)) |
446 | return -ENODEV; | |
3ef28e83 DW |
447 | } |
448 | } | |
449 | ||
450 | void blk_queue_exit(struct request_queue *q) | |
451 | { | |
452 | percpu_ref_put(&q->q_usage_counter); | |
453 | } | |
454 | ||
455 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
456 | { | |
457 | struct request_queue *q = | |
458 | container_of(ref, struct request_queue, q_usage_counter); | |
459 | ||
460 | wake_up_all(&q->mq_freeze_wq); | |
461 | } | |
462 | ||
bca237a5 | 463 | static void blk_rq_timed_out_timer(struct timer_list *t) |
287922eb | 464 | { |
bca237a5 | 465 | struct request_queue *q = from_timer(q, t, timeout); |
287922eb CH |
466 | |
467 | kblockd_schedule_work(&q->timeout_work); | |
468 | } | |
469 | ||
2e3c18d0 TH |
470 | static void blk_timeout_work(struct work_struct *work) |
471 | { | |
472 | } | |
473 | ||
498f6650 BVA |
474 | /** |
475 | * blk_alloc_queue_node - allocate a request queue | |
476 | * @gfp_mask: memory allocation flags | |
477 | * @node_id: NUMA node to allocate memory from | |
498f6650 | 478 | */ |
6d469642 | 479 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 480 | { |
165125e1 | 481 | struct request_queue *q; |
338aa96d | 482 | int ret; |
1946089a | 483 | |
8324aa91 | 484 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 485 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
486 | if (!q) |
487 | return NULL; | |
488 | ||
cbf62af3 | 489 | q->last_merge = NULL; |
cbf62af3 | 490 | |
00380a40 | 491 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d | 492 | if (q->id < 0) |
3d2936f4 | 493 | goto fail_q; |
a73f730d | 494 | |
338aa96d KO |
495 | ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
496 | if (ret) | |
54efd50b KO |
497 | goto fail_id; |
498 | ||
d03f6cdc JK |
499 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); |
500 | if (!q->backing_dev_info) | |
501 | goto fail_split; | |
502 | ||
a83b576c JA |
503 | q->stats = blk_alloc_queue_stats(); |
504 | if (!q->stats) | |
505 | goto fail_stats; | |
506 | ||
b5420237 | 507 | q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES; |
dc3b17cc JK |
508 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; |
509 | q->backing_dev_info->name = "block"; | |
5151412d | 510 | q->node = node_id; |
0989a025 | 511 | |
bca237a5 KC |
512 | timer_setup(&q->backing_dev_info->laptop_mode_wb_timer, |
513 | laptop_mode_timer_fn, 0); | |
514 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); | |
2e3c18d0 | 515 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
a612fddf | 516 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 517 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 518 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 519 | #endif |
483f4afc | 520 | |
8324aa91 | 521 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 522 | |
5acb3cc2 WL |
523 | #ifdef CONFIG_BLK_DEV_IO_TRACE |
524 | mutex_init(&q->blk_trace_mutex); | |
525 | #endif | |
483f4afc | 526 | mutex_init(&q->sysfs_lock); |
cecf5d87 | 527 | mutex_init(&q->sysfs_dir_lock); |
0d945c1f | 528 | spin_lock_init(&q->queue_lock); |
c94a96ac | 529 | |
320ae51f | 530 | init_waitqueue_head(&q->mq_freeze_wq); |
7996a8b5 | 531 | mutex_init(&q->mq_freeze_lock); |
320ae51f | 532 | |
3ef28e83 DW |
533 | /* |
534 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
535 | * See blk_register_queue() for details. | |
536 | */ | |
537 | if (percpu_ref_init(&q->q_usage_counter, | |
538 | blk_queue_usage_counter_release, | |
539 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 540 | goto fail_bdi; |
f51b802c | 541 | |
3ef28e83 DW |
542 | if (blkcg_init_queue(q)) |
543 | goto fail_ref; | |
544 | ||
1da177e4 | 545 | return q; |
a73f730d | 546 | |
3ef28e83 DW |
547 | fail_ref: |
548 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b | 549 | fail_bdi: |
a83b576c JA |
550 | blk_free_queue_stats(q->stats); |
551 | fail_stats: | |
d03f6cdc | 552 | bdi_put(q->backing_dev_info); |
54efd50b | 553 | fail_split: |
338aa96d | 554 | bioset_exit(&q->bio_split); |
a73f730d TH |
555 | fail_id: |
556 | ida_simple_remove(&blk_queue_ida, q->id); | |
557 | fail_q: | |
558 | kmem_cache_free(blk_requestq_cachep, q); | |
559 | return NULL; | |
1da177e4 | 560 | } |
1946089a | 561 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 | 562 | |
09ac46c4 | 563 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 564 | { |
3f3299d5 | 565 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
566 | __blk_get_queue(q); |
567 | return true; | |
1da177e4 LT |
568 | } |
569 | ||
09ac46c4 | 570 | return false; |
1da177e4 | 571 | } |
d86e0e83 | 572 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 573 | |
a1ce35fa JA |
574 | /** |
575 | * blk_get_request - allocate a request | |
576 | * @q: request queue to allocate a request for | |
577 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. | |
578 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. | |
1da177e4 | 579 | */ |
a1ce35fa JA |
580 | struct request *blk_get_request(struct request_queue *q, unsigned int op, |
581 | blk_mq_req_flags_t flags) | |
1da177e4 | 582 | { |
a1ce35fa | 583 | struct request *req; |
1da177e4 | 584 | |
a1ce35fa JA |
585 | WARN_ON_ONCE(op & REQ_NOWAIT); |
586 | WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT)); | |
1da177e4 | 587 | |
a1ce35fa JA |
588 | req = blk_mq_alloc_request(q, op, flags); |
589 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) | |
590 | q->mq_ops->initialize_rq_fn(req); | |
1da177e4 | 591 | |
a1ce35fa | 592 | return req; |
1da177e4 | 593 | } |
a1ce35fa | 594 | EXPORT_SYMBOL(blk_get_request); |
1da177e4 | 595 | |
1da177e4 LT |
596 | void blk_put_request(struct request *req) |
597 | { | |
a1ce35fa | 598 | blk_mq_free_request(req); |
1da177e4 | 599 | } |
1da177e4 LT |
600 | EXPORT_SYMBOL(blk_put_request); |
601 | ||
14ccb66b CH |
602 | bool bio_attempt_back_merge(struct request *req, struct bio *bio, |
603 | unsigned int nr_segs) | |
73c10101 | 604 | { |
1eff9d32 | 605 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 606 | |
14ccb66b | 607 | if (!ll_back_merge_fn(req, bio, nr_segs)) |
73c10101 JA |
608 | return false; |
609 | ||
14ccb66b | 610 | trace_block_bio_backmerge(req->q, req, bio); |
d3e65fff | 611 | rq_qos_merge(req->q, req, bio); |
73c10101 JA |
612 | |
613 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
614 | blk_rq_set_mixed_merge(req); | |
615 | ||
616 | req->biotail->bi_next = bio; | |
617 | req->biotail = bio; | |
4f024f37 | 618 | req->__data_len += bio->bi_iter.bi_size; |
73c10101 | 619 | |
320ae51f | 620 | blk_account_io_start(req, false); |
73c10101 JA |
621 | return true; |
622 | } | |
623 | ||
14ccb66b CH |
624 | bool bio_attempt_front_merge(struct request *req, struct bio *bio, |
625 | unsigned int nr_segs) | |
73c10101 | 626 | { |
1eff9d32 | 627 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 628 | |
14ccb66b | 629 | if (!ll_front_merge_fn(req, bio, nr_segs)) |
73c10101 JA |
630 | return false; |
631 | ||
14ccb66b | 632 | trace_block_bio_frontmerge(req->q, req, bio); |
d3e65fff | 633 | rq_qos_merge(req->q, req, bio); |
73c10101 JA |
634 | |
635 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
636 | blk_rq_set_mixed_merge(req); | |
637 | ||
73c10101 JA |
638 | bio->bi_next = req->bio; |
639 | req->bio = bio; | |
640 | ||
4f024f37 KO |
641 | req->__sector = bio->bi_iter.bi_sector; |
642 | req->__data_len += bio->bi_iter.bi_size; | |
73c10101 | 643 | |
320ae51f | 644 | blk_account_io_start(req, false); |
73c10101 JA |
645 | return true; |
646 | } | |
647 | ||
1e739730 CH |
648 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, |
649 | struct bio *bio) | |
650 | { | |
651 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
652 | ||
653 | if (segments >= queue_max_discard_segments(q)) | |
654 | goto no_merge; | |
655 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
656 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
657 | goto no_merge; | |
658 | ||
d3e65fff TH |
659 | rq_qos_merge(q, req, bio); |
660 | ||
1e739730 CH |
661 | req->biotail->bi_next = bio; |
662 | req->biotail = bio; | |
663 | req->__data_len += bio->bi_iter.bi_size; | |
1e739730 CH |
664 | req->nr_phys_segments = segments + 1; |
665 | ||
666 | blk_account_io_start(req, false); | |
667 | return true; | |
668 | no_merge: | |
669 | req_set_nomerge(q, req); | |
670 | return false; | |
671 | } | |
672 | ||
bd87b589 | 673 | /** |
320ae51f | 674 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
bd87b589 TH |
675 | * @q: request_queue new bio is being queued at |
676 | * @bio: new bio being queued | |
14ccb66b | 677 | * @nr_segs: number of segments in @bio |
ccc2600b RD |
678 | * @same_queue_rq: pointer to &struct request that gets filled in when |
679 | * another request associated with @q is found on the plug list | |
680 | * (optional, may be %NULL) | |
bd87b589 TH |
681 | * |
682 | * Determine whether @bio being queued on @q can be merged with a request | |
683 | * on %current's plugged list. Returns %true if merge was successful, | |
684 | * otherwise %false. | |
685 | * | |
07c2bd37 TH |
686 | * Plugging coalesces IOs from the same issuer for the same purpose without |
687 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
688 | * than scheduling, and the request, while may have elvpriv data, is not | |
689 | * added on the elevator at this point. In addition, we don't have | |
690 | * reliable access to the elevator outside queue lock. Only check basic | |
691 | * merging parameters without querying the elevator. | |
da41a589 RE |
692 | * |
693 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
73c10101 | 694 | */ |
320ae51f | 695 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
14ccb66b | 696 | unsigned int nr_segs, struct request **same_queue_rq) |
73c10101 JA |
697 | { |
698 | struct blk_plug *plug; | |
699 | struct request *rq; | |
92f399c7 | 700 | struct list_head *plug_list; |
73c10101 | 701 | |
b49773e7 | 702 | plug = blk_mq_plug(q, bio); |
73c10101 | 703 | if (!plug) |
34fe7c05 | 704 | return false; |
73c10101 | 705 | |
a1ce35fa | 706 | plug_list = &plug->mq_list; |
92f399c7 SL |
707 | |
708 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
34fe7c05 | 709 | bool merged = false; |
73c10101 | 710 | |
5f0ed774 | 711 | if (rq->q == q && same_queue_rq) { |
5b3f341f SL |
712 | /* |
713 | * Only blk-mq multiple hardware queues case checks the | |
714 | * rq in the same queue, there should be only one such | |
715 | * rq in a queue | |
716 | **/ | |
5f0ed774 | 717 | *same_queue_rq = rq; |
5b3f341f | 718 | } |
56ebdaf2 | 719 | |
07c2bd37 | 720 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
721 | continue; |
722 | ||
34fe7c05 CH |
723 | switch (blk_try_merge(rq, bio)) { |
724 | case ELEVATOR_BACK_MERGE: | |
14ccb66b | 725 | merged = bio_attempt_back_merge(rq, bio, nr_segs); |
34fe7c05 CH |
726 | break; |
727 | case ELEVATOR_FRONT_MERGE: | |
14ccb66b | 728 | merged = bio_attempt_front_merge(rq, bio, nr_segs); |
34fe7c05 | 729 | break; |
1e739730 CH |
730 | case ELEVATOR_DISCARD_MERGE: |
731 | merged = bio_attempt_discard_merge(q, rq, bio); | |
732 | break; | |
34fe7c05 CH |
733 | default: |
734 | break; | |
73c10101 | 735 | } |
34fe7c05 CH |
736 | |
737 | if (merged) | |
738 | return true; | |
73c10101 | 739 | } |
34fe7c05 CH |
740 | |
741 | return false; | |
73c10101 JA |
742 | } |
743 | ||
52c5e62d | 744 | static void handle_bad_sector(struct bio *bio, sector_t maxsector) |
1da177e4 LT |
745 | { |
746 | char b[BDEVNAME_SIZE]; | |
747 | ||
748 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
6296b960 | 749 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", |
74d46992 | 750 | bio_devname(bio, b), bio->bi_opf, |
f73a1c7d | 751 | (unsigned long long)bio_end_sector(bio), |
52c5e62d | 752 | (long long)maxsector); |
1da177e4 LT |
753 | } |
754 | ||
c17bb495 AM |
755 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
756 | ||
757 | static DECLARE_FAULT_ATTR(fail_make_request); | |
758 | ||
759 | static int __init setup_fail_make_request(char *str) | |
760 | { | |
761 | return setup_fault_attr(&fail_make_request, str); | |
762 | } | |
763 | __setup("fail_make_request=", setup_fail_make_request); | |
764 | ||
b2c9cd37 | 765 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 766 | { |
b2c9cd37 | 767 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
768 | } |
769 | ||
770 | static int __init fail_make_request_debugfs(void) | |
771 | { | |
dd48c085 AM |
772 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
773 | NULL, &fail_make_request); | |
774 | ||
21f9fcd8 | 775 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
776 | } |
777 | ||
778 | late_initcall(fail_make_request_debugfs); | |
779 | ||
780 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
781 | ||
b2c9cd37 AM |
782 | static inline bool should_fail_request(struct hd_struct *part, |
783 | unsigned int bytes) | |
c17bb495 | 784 | { |
b2c9cd37 | 785 | return false; |
c17bb495 AM |
786 | } |
787 | ||
788 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
789 | ||
721c7fc7 ID |
790 | static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part) |
791 | { | |
b089cfd9 JA |
792 | const int op = bio_op(bio); |
793 | ||
8b2ded1c | 794 | if (part->policy && op_is_write(op)) { |
721c7fc7 ID |
795 | char b[BDEVNAME_SIZE]; |
796 | ||
8b2ded1c MP |
797 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) |
798 | return false; | |
799 | ||
a32e236e | 800 | WARN_ONCE(1, |
721c7fc7 ID |
801 | "generic_make_request: Trying to write " |
802 | "to read-only block-device %s (partno %d)\n", | |
803 | bio_devname(bio, b), part->partno); | |
a32e236e LT |
804 | /* Older lvm-tools actually trigger this */ |
805 | return false; | |
721c7fc7 ID |
806 | } |
807 | ||
808 | return false; | |
809 | } | |
810 | ||
30abb3a6 HM |
811 | static noinline int should_fail_bio(struct bio *bio) |
812 | { | |
813 | if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) | |
814 | return -EIO; | |
815 | return 0; | |
816 | } | |
817 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
818 | ||
52c5e62d CH |
819 | /* |
820 | * Check whether this bio extends beyond the end of the device or partition. | |
821 | * This may well happen - the kernel calls bread() without checking the size of | |
822 | * the device, e.g., when mounting a file system. | |
823 | */ | |
824 | static inline int bio_check_eod(struct bio *bio, sector_t maxsector) | |
825 | { | |
826 | unsigned int nr_sectors = bio_sectors(bio); | |
827 | ||
828 | if (nr_sectors && maxsector && | |
829 | (nr_sectors > maxsector || | |
830 | bio->bi_iter.bi_sector > maxsector - nr_sectors)) { | |
831 | handle_bad_sector(bio, maxsector); | |
832 | return -EIO; | |
833 | } | |
834 | return 0; | |
835 | } | |
836 | ||
74d46992 CH |
837 | /* |
838 | * Remap block n of partition p to block n+start(p) of the disk. | |
839 | */ | |
840 | static inline int blk_partition_remap(struct bio *bio) | |
841 | { | |
842 | struct hd_struct *p; | |
52c5e62d | 843 | int ret = -EIO; |
74d46992 | 844 | |
721c7fc7 ID |
845 | rcu_read_lock(); |
846 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
52c5e62d CH |
847 | if (unlikely(!p)) |
848 | goto out; | |
849 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) | |
850 | goto out; | |
851 | if (unlikely(bio_check_ro(bio, p))) | |
721c7fc7 | 852 | goto out; |
721c7fc7 | 853 | |
74d46992 | 854 | /* |
6c1b1da5 AJ |
855 | * Zone management bios do not have a sector count but they do have |
856 | * a start sector filled out and need to be remapped. | |
74d46992 | 857 | */ |
6c1b1da5 | 858 | if (bio_sectors(bio) || op_is_zone_mgmt(bio_op(bio))) { |
52c5e62d CH |
859 | if (bio_check_eod(bio, part_nr_sects_read(p))) |
860 | goto out; | |
861 | bio->bi_iter.bi_sector += p->start_sect; | |
52c5e62d CH |
862 | trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), |
863 | bio->bi_iter.bi_sector - p->start_sect); | |
864 | } | |
c04fa44b | 865 | bio->bi_partno = 0; |
52c5e62d | 866 | ret = 0; |
721c7fc7 ID |
867 | out: |
868 | rcu_read_unlock(); | |
74d46992 CH |
869 | return ret; |
870 | } | |
871 | ||
27a84d54 CH |
872 | static noinline_for_stack bool |
873 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 874 | { |
165125e1 | 875 | struct request_queue *q; |
5a7bbad2 | 876 | int nr_sectors = bio_sectors(bio); |
4e4cbee9 | 877 | blk_status_t status = BLK_STS_IOERR; |
5a7bbad2 | 878 | char b[BDEVNAME_SIZE]; |
1da177e4 LT |
879 | |
880 | might_sleep(); | |
1da177e4 | 881 | |
74d46992 | 882 | q = bio->bi_disk->queue; |
5a7bbad2 CH |
883 | if (unlikely(!q)) { |
884 | printk(KERN_ERR | |
885 | "generic_make_request: Trying to access " | |
886 | "nonexistent block-device %s (%Lu)\n", | |
74d46992 | 887 | bio_devname(bio, b), (long long)bio->bi_iter.bi_sector); |
5a7bbad2 CH |
888 | goto end_io; |
889 | } | |
c17bb495 | 890 | |
03a07c92 GR |
891 | /* |
892 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP | |
893 | * if queue is not a request based queue. | |
894 | */ | |
344e9ffc | 895 | if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q)) |
03a07c92 GR |
896 | goto not_supported; |
897 | ||
30abb3a6 | 898 | if (should_fail_bio(bio)) |
5a7bbad2 | 899 | goto end_io; |
2056a782 | 900 | |
52c5e62d CH |
901 | if (bio->bi_partno) { |
902 | if (unlikely(blk_partition_remap(bio))) | |
721c7fc7 ID |
903 | goto end_io; |
904 | } else { | |
52c5e62d CH |
905 | if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0))) |
906 | goto end_io; | |
907 | if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk)))) | |
721c7fc7 ID |
908 | goto end_io; |
909 | } | |
2056a782 | 910 | |
5a7bbad2 CH |
911 | /* |
912 | * Filter flush bio's early so that make_request based | |
913 | * drivers without flush support don't have to worry | |
914 | * about them. | |
915 | */ | |
f3a8ab7d | 916 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 917 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 918 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
5a7bbad2 | 919 | if (!nr_sectors) { |
4e4cbee9 | 920 | status = BLK_STS_OK; |
51fd77bd JA |
921 | goto end_io; |
922 | } | |
5a7bbad2 | 923 | } |
5ddfe969 | 924 | |
d04c406f CH |
925 | if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) |
926 | bio->bi_opf &= ~REQ_HIPRI; | |
927 | ||
288dab8a CH |
928 | switch (bio_op(bio)) { |
929 | case REQ_OP_DISCARD: | |
930 | if (!blk_queue_discard(q)) | |
931 | goto not_supported; | |
932 | break; | |
933 | case REQ_OP_SECURE_ERASE: | |
934 | if (!blk_queue_secure_erase(q)) | |
935 | goto not_supported; | |
936 | break; | |
937 | case REQ_OP_WRITE_SAME: | |
74d46992 | 938 | if (!q->limits.max_write_same_sectors) |
288dab8a | 939 | goto not_supported; |
58886785 | 940 | break; |
2d253440 | 941 | case REQ_OP_ZONE_RESET: |
6c1b1da5 AJ |
942 | case REQ_OP_ZONE_OPEN: |
943 | case REQ_OP_ZONE_CLOSE: | |
944 | case REQ_OP_ZONE_FINISH: | |
74d46992 | 945 | if (!blk_queue_is_zoned(q)) |
2d253440 | 946 | goto not_supported; |
288dab8a | 947 | break; |
6e33dbf2 CK |
948 | case REQ_OP_ZONE_RESET_ALL: |
949 | if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q)) | |
950 | goto not_supported; | |
951 | break; | |
a6f0788e | 952 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 953 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
954 | goto not_supported; |
955 | break; | |
288dab8a CH |
956 | default: |
957 | break; | |
5a7bbad2 | 958 | } |
01edede4 | 959 | |
7f4b35d1 TH |
960 | /* |
961 | * Various block parts want %current->io_context and lazy ioc | |
962 | * allocation ends up trading a lot of pain for a small amount of | |
963 | * memory. Just allocate it upfront. This may fail and block | |
964 | * layer knows how to live with it. | |
965 | */ | |
966 | create_io_context(GFP_ATOMIC, q->node); | |
967 | ||
ae118896 TH |
968 | if (!blkcg_bio_issue_check(q, bio)) |
969 | return false; | |
27a84d54 | 970 | |
fbbaf700 N |
971 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
972 | trace_block_bio_queue(q, bio); | |
973 | /* Now that enqueuing has been traced, we need to trace | |
974 | * completion as well. | |
975 | */ | |
976 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
977 | } | |
27a84d54 | 978 | return true; |
a7384677 | 979 | |
288dab8a | 980 | not_supported: |
4e4cbee9 | 981 | status = BLK_STS_NOTSUPP; |
a7384677 | 982 | end_io: |
4e4cbee9 | 983 | bio->bi_status = status; |
4246a0b6 | 984 | bio_endio(bio); |
27a84d54 | 985 | return false; |
1da177e4 LT |
986 | } |
987 | ||
27a84d54 CH |
988 | /** |
989 | * generic_make_request - hand a buffer to its device driver for I/O | |
990 | * @bio: The bio describing the location in memory and on the device. | |
991 | * | |
992 | * generic_make_request() is used to make I/O requests of block | |
993 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
994 | * to be done. | |
995 | * | |
996 | * generic_make_request() does not return any status. The | |
997 | * success/failure status of the request, along with notification of | |
998 | * completion, is delivered asynchronously through the bio->bi_end_io | |
999 | * function described (one day) else where. | |
1000 | * | |
1001 | * The caller of generic_make_request must make sure that bi_io_vec | |
1002 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1003 | * set to describe the device address, and the | |
1004 | * bi_end_io and optionally bi_private are set to describe how | |
1005 | * completion notification should be signaled. | |
1006 | * | |
1007 | * generic_make_request and the drivers it calls may use bi_next if this | |
1008 | * bio happens to be merged with someone else, and may resubmit the bio to | |
1009 | * a lower device by calling into generic_make_request recursively, which | |
1010 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 | 1011 | */ |
dece1635 | 1012 | blk_qc_t generic_make_request(struct bio *bio) |
d89d8796 | 1013 | { |
f5fe1b51 N |
1014 | /* |
1015 | * bio_list_on_stack[0] contains bios submitted by the current | |
1016 | * make_request_fn. | |
1017 | * bio_list_on_stack[1] contains bios that were submitted before | |
1018 | * the current make_request_fn, but that haven't been processed | |
1019 | * yet. | |
1020 | */ | |
1021 | struct bio_list bio_list_on_stack[2]; | |
dece1635 | 1022 | blk_qc_t ret = BLK_QC_T_NONE; |
bddd87c7 | 1023 | |
27a84d54 | 1024 | if (!generic_make_request_checks(bio)) |
dece1635 | 1025 | goto out; |
27a84d54 CH |
1026 | |
1027 | /* | |
1028 | * We only want one ->make_request_fn to be active at a time, else | |
1029 | * stack usage with stacked devices could be a problem. So use | |
1030 | * current->bio_list to keep a list of requests submited by a | |
1031 | * make_request_fn function. current->bio_list is also used as a | |
1032 | * flag to say if generic_make_request is currently active in this | |
1033 | * task or not. If it is NULL, then no make_request is active. If | |
1034 | * it is non-NULL, then a make_request is active, and new requests | |
1035 | * should be added at the tail | |
1036 | */ | |
bddd87c7 | 1037 | if (current->bio_list) { |
f5fe1b51 | 1038 | bio_list_add(¤t->bio_list[0], bio); |
dece1635 | 1039 | goto out; |
d89d8796 | 1040 | } |
27a84d54 | 1041 | |
d89d8796 NB |
1042 | /* following loop may be a bit non-obvious, and so deserves some |
1043 | * explanation. | |
1044 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1045 | * ensure that) so we have a list with a single bio. | |
1046 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1047 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1048 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 1049 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
1050 | * through a recursive call to generic_make_request. If it |
1051 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1052 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 1053 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 1054 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
1055 | */ |
1056 | BUG_ON(bio->bi_next); | |
f5fe1b51 N |
1057 | bio_list_init(&bio_list_on_stack[0]); |
1058 | current->bio_list = bio_list_on_stack; | |
d89d8796 | 1059 | do { |
fe200864 ML |
1060 | struct request_queue *q = bio->bi_disk->queue; |
1061 | blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ? | |
1062 | BLK_MQ_REQ_NOWAIT : 0; | |
27a84d54 | 1063 | |
fe200864 | 1064 | if (likely(blk_queue_enter(q, flags) == 0)) { |
79bd9959 N |
1065 | struct bio_list lower, same; |
1066 | ||
1067 | /* Create a fresh bio_list for all subordinate requests */ | |
f5fe1b51 N |
1068 | bio_list_on_stack[1] = bio_list_on_stack[0]; |
1069 | bio_list_init(&bio_list_on_stack[0]); | |
dece1635 | 1070 | ret = q->make_request_fn(q, bio); |
3ef28e83 | 1071 | |
fe200864 ML |
1072 | blk_queue_exit(q); |
1073 | ||
79bd9959 N |
1074 | /* sort new bios into those for a lower level |
1075 | * and those for the same level | |
1076 | */ | |
1077 | bio_list_init(&lower); | |
1078 | bio_list_init(&same); | |
f5fe1b51 | 1079 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) |
74d46992 | 1080 | if (q == bio->bi_disk->queue) |
79bd9959 N |
1081 | bio_list_add(&same, bio); |
1082 | else | |
1083 | bio_list_add(&lower, bio); | |
1084 | /* now assemble so we handle the lowest level first */ | |
f5fe1b51 N |
1085 | bio_list_merge(&bio_list_on_stack[0], &lower); |
1086 | bio_list_merge(&bio_list_on_stack[0], &same); | |
1087 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
3ef28e83 | 1088 | } else { |
03a07c92 GR |
1089 | if (unlikely(!blk_queue_dying(q) && |
1090 | (bio->bi_opf & REQ_NOWAIT))) | |
1091 | bio_wouldblock_error(bio); | |
1092 | else | |
1093 | bio_io_error(bio); | |
3ef28e83 | 1094 | } |
f5fe1b51 | 1095 | bio = bio_list_pop(&bio_list_on_stack[0]); |
d89d8796 | 1096 | } while (bio); |
bddd87c7 | 1097 | current->bio_list = NULL; /* deactivate */ |
dece1635 JA |
1098 | |
1099 | out: | |
1100 | return ret; | |
d89d8796 | 1101 | } |
1da177e4 LT |
1102 | EXPORT_SYMBOL(generic_make_request); |
1103 | ||
f421e1d9 CH |
1104 | /** |
1105 | * direct_make_request - hand a buffer directly to its device driver for I/O | |
1106 | * @bio: The bio describing the location in memory and on the device. | |
1107 | * | |
1108 | * This function behaves like generic_make_request(), but does not protect | |
1109 | * against recursion. Must only be used if the called driver is known | |
1110 | * to not call generic_make_request (or direct_make_request) again from | |
1111 | * its make_request function. (Calling direct_make_request again from | |
1112 | * a workqueue is perfectly fine as that doesn't recurse). | |
1113 | */ | |
1114 | blk_qc_t direct_make_request(struct bio *bio) | |
1115 | { | |
1116 | struct request_queue *q = bio->bi_disk->queue; | |
1117 | bool nowait = bio->bi_opf & REQ_NOWAIT; | |
1118 | blk_qc_t ret; | |
1119 | ||
1120 | if (!generic_make_request_checks(bio)) | |
1121 | return BLK_QC_T_NONE; | |
1122 | ||
3a0a5299 | 1123 | if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) { |
f421e1d9 CH |
1124 | if (nowait && !blk_queue_dying(q)) |
1125 | bio->bi_status = BLK_STS_AGAIN; | |
1126 | else | |
1127 | bio->bi_status = BLK_STS_IOERR; | |
1128 | bio_endio(bio); | |
1129 | return BLK_QC_T_NONE; | |
1130 | } | |
1131 | ||
1132 | ret = q->make_request_fn(q, bio); | |
1133 | blk_queue_exit(q); | |
1134 | return ret; | |
1135 | } | |
1136 | EXPORT_SYMBOL_GPL(direct_make_request); | |
1137 | ||
1da177e4 | 1138 | /** |
710027a4 | 1139 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1140 | * @bio: The &struct bio which describes the I/O |
1141 | * | |
1142 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1143 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1144 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1145 | * |
1146 | */ | |
4e49ea4a | 1147 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 1148 | { |
b8e24a93 JW |
1149 | bool workingset_read = false; |
1150 | unsigned long pflags; | |
1151 | blk_qc_t ret; | |
1152 | ||
d3f77dfd TH |
1153 | if (blkcg_punt_bio_submit(bio)) |
1154 | return BLK_QC_T_NONE; | |
1155 | ||
bf2de6f5 JA |
1156 | /* |
1157 | * If it's a regular read/write or a barrier with data attached, | |
1158 | * go through the normal accounting stuff before submission. | |
1159 | */ | |
e2a60da7 | 1160 | if (bio_has_data(bio)) { |
4363ac7c MP |
1161 | unsigned int count; |
1162 | ||
95fe6c1a | 1163 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
7c5a0dcf | 1164 | count = queue_logical_block_size(bio->bi_disk->queue) >> 9; |
4363ac7c MP |
1165 | else |
1166 | count = bio_sectors(bio); | |
1167 | ||
a8ebb056 | 1168 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
1169 | count_vm_events(PGPGOUT, count); |
1170 | } else { | |
b8e24a93 JW |
1171 | if (bio_flagged(bio, BIO_WORKINGSET)) |
1172 | workingset_read = true; | |
4f024f37 | 1173 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
1174 | count_vm_events(PGPGIN, count); |
1175 | } | |
1176 | ||
1177 | if (unlikely(block_dump)) { | |
1178 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1179 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1180 | current->comm, task_pid_nr(current), |
a8ebb056 | 1181 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 1182 | (unsigned long long)bio->bi_iter.bi_sector, |
74d46992 | 1183 | bio_devname(bio, b), count); |
bf2de6f5 | 1184 | } |
1da177e4 LT |
1185 | } |
1186 | ||
b8e24a93 JW |
1187 | /* |
1188 | * If we're reading data that is part of the userspace | |
1189 | * workingset, count submission time as memory stall. When the | |
1190 | * device is congested, or the submitting cgroup IO-throttled, | |
1191 | * submission can be a significant part of overall IO time. | |
1192 | */ | |
1193 | if (workingset_read) | |
1194 | psi_memstall_enter(&pflags); | |
1195 | ||
1196 | ret = generic_make_request(bio); | |
1197 | ||
1198 | if (workingset_read) | |
1199 | psi_memstall_leave(&pflags); | |
1200 | ||
1201 | return ret; | |
1da177e4 | 1202 | } |
1da177e4 LT |
1203 | EXPORT_SYMBOL(submit_bio); |
1204 | ||
82124d60 | 1205 | /** |
bf4e6b4e HR |
1206 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
1207 | * for new the queue limits | |
82124d60 KU |
1208 | * @q: the queue |
1209 | * @rq: the request being checked | |
1210 | * | |
1211 | * Description: | |
1212 | * @rq may have been made based on weaker limitations of upper-level queues | |
1213 | * in request stacking drivers, and it may violate the limitation of @q. | |
1214 | * Since the block layer and the underlying device driver trust @rq | |
1215 | * after it is inserted to @q, it should be checked against @q before | |
1216 | * the insertion using this generic function. | |
1217 | * | |
82124d60 | 1218 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
1219 | * limits when retrying requests on other queues. Those requests need |
1220 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 1221 | */ |
bf4e6b4e HR |
1222 | static int blk_cloned_rq_check_limits(struct request_queue *q, |
1223 | struct request *rq) | |
82124d60 | 1224 | { |
8fe0d473 | 1225 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { |
61939b12 JP |
1226 | printk(KERN_ERR "%s: over max size limit. (%u > %u)\n", |
1227 | __func__, blk_rq_sectors(rq), | |
1228 | blk_queue_get_max_sectors(q, req_op(rq))); | |
82124d60 KU |
1229 | return -EIO; |
1230 | } | |
1231 | ||
1232 | /* | |
1233 | * queue's settings related to segment counting like q->bounce_pfn | |
1234 | * may differ from that of other stacking queues. | |
1235 | * Recalculate it to check the request correctly on this queue's | |
1236 | * limitation. | |
1237 | */ | |
e9cd19c0 | 1238 | rq->nr_phys_segments = blk_recalc_rq_segments(rq); |
8a78362c | 1239 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
61939b12 JP |
1240 | printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n", |
1241 | __func__, rq->nr_phys_segments, queue_max_segments(q)); | |
82124d60 KU |
1242 | return -EIO; |
1243 | } | |
1244 | ||
1245 | return 0; | |
1246 | } | |
82124d60 KU |
1247 | |
1248 | /** | |
1249 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1250 | * @q: the queue to submit the request | |
1251 | * @rq: the request being queued | |
1252 | */ | |
2a842aca | 1253 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
82124d60 | 1254 | { |
bf4e6b4e | 1255 | if (blk_cloned_rq_check_limits(q, rq)) |
2a842aca | 1256 | return BLK_STS_IOERR; |
82124d60 | 1257 | |
b2c9cd37 AM |
1258 | if (rq->rq_disk && |
1259 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
2a842aca | 1260 | return BLK_STS_IOERR; |
82124d60 | 1261 | |
a1ce35fa JA |
1262 | if (blk_queue_io_stat(q)) |
1263 | blk_account_io_start(rq, true); | |
82124d60 KU |
1264 | |
1265 | /* | |
a1ce35fa JA |
1266 | * Since we have a scheduler attached on the top device, |
1267 | * bypass a potential scheduler on the bottom device for | |
1268 | * insert. | |
82124d60 | 1269 | */ |
fd9c40f6 | 1270 | return blk_mq_request_issue_directly(rq, true); |
82124d60 KU |
1271 | } |
1272 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1273 | ||
80a761fd TH |
1274 | /** |
1275 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1276 | * @rq: request to examine | |
1277 | * | |
1278 | * Description: | |
1279 | * A request could be merge of IOs which require different failure | |
1280 | * handling. This function determines the number of bytes which | |
1281 | * can be failed from the beginning of the request without | |
1282 | * crossing into area which need to be retried further. | |
1283 | * | |
1284 | * Return: | |
1285 | * The number of bytes to fail. | |
80a761fd TH |
1286 | */ |
1287 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1288 | { | |
1289 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1290 | unsigned int bytes = 0; | |
1291 | struct bio *bio; | |
1292 | ||
e8064021 | 1293 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
1294 | return blk_rq_bytes(rq); |
1295 | ||
1296 | /* | |
1297 | * Currently the only 'mixing' which can happen is between | |
1298 | * different fastfail types. We can safely fail portions | |
1299 | * which have all the failfast bits that the first one has - | |
1300 | * the ones which are at least as eager to fail as the first | |
1301 | * one. | |
1302 | */ | |
1303 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 1304 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 1305 | break; |
4f024f37 | 1306 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
1307 | } |
1308 | ||
1309 | /* this could lead to infinite loop */ | |
1310 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1311 | return bytes; | |
1312 | } | |
1313 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1314 | ||
320ae51f | 1315 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 1316 | { |
c2553b58 | 1317 | if (blk_do_io_stat(req)) { |
ddcf35d3 | 1318 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 | 1319 | struct hd_struct *part; |
bc58ba94 | 1320 | |
112f158f | 1321 | part_stat_lock(); |
09e099d4 | 1322 | part = req->part; |
112f158f | 1323 | part_stat_add(part, sectors[sgrp], bytes >> 9); |
bc58ba94 JA |
1324 | part_stat_unlock(); |
1325 | } | |
1326 | } | |
1327 | ||
522a7775 | 1328 | void blk_account_io_done(struct request *req, u64 now) |
bc58ba94 | 1329 | { |
bc58ba94 | 1330 | /* |
dd4c133f TH |
1331 | * Account IO completion. flush_rq isn't accounted as a |
1332 | * normal IO on queueing nor completion. Accounting the | |
1333 | * containing request is enough. | |
bc58ba94 | 1334 | */ |
e8064021 | 1335 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { |
ddcf35d3 | 1336 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 | 1337 | struct hd_struct *part; |
bc58ba94 | 1338 | |
112f158f | 1339 | part_stat_lock(); |
09e099d4 | 1340 | part = req->part; |
bc58ba94 | 1341 | |
5b18b5a7 | 1342 | update_io_ticks(part, jiffies); |
112f158f MS |
1343 | part_stat_inc(part, ios[sgrp]); |
1344 | part_stat_add(part, nsecs[sgrp], now - req->start_time_ns); | |
5b18b5a7 | 1345 | part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns)); |
ddcf35d3 | 1346 | part_dec_in_flight(req->q, part, rq_data_dir(req)); |
bc58ba94 | 1347 | |
6c23a968 | 1348 | hd_struct_put(part); |
bc58ba94 JA |
1349 | part_stat_unlock(); |
1350 | } | |
1351 | } | |
1352 | ||
320ae51f JA |
1353 | void blk_account_io_start(struct request *rq, bool new_io) |
1354 | { | |
1355 | struct hd_struct *part; | |
1356 | int rw = rq_data_dir(rq); | |
320ae51f JA |
1357 | |
1358 | if (!blk_do_io_stat(rq)) | |
1359 | return; | |
1360 | ||
112f158f | 1361 | part_stat_lock(); |
320ae51f JA |
1362 | |
1363 | if (!new_io) { | |
1364 | part = rq->part; | |
112f158f | 1365 | part_stat_inc(part, merges[rw]); |
320ae51f JA |
1366 | } else { |
1367 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
1368 | if (!hd_struct_try_get(part)) { | |
1369 | /* | |
1370 | * The partition is already being removed, | |
1371 | * the request will be accounted on the disk only | |
1372 | * | |
1373 | * We take a reference on disk->part0 although that | |
1374 | * partition will never be deleted, so we can treat | |
1375 | * it as any other partition. | |
1376 | */ | |
1377 | part = &rq->rq_disk->part0; | |
1378 | hd_struct_get(part); | |
1379 | } | |
d62e26b3 | 1380 | part_inc_in_flight(rq->q, part, rw); |
320ae51f JA |
1381 | rq->part = part; |
1382 | } | |
1383 | ||
5b18b5a7 MP |
1384 | update_io_ticks(part, jiffies); |
1385 | ||
320ae51f JA |
1386 | part_stat_unlock(); |
1387 | } | |
1388 | ||
ef71de8b CH |
1389 | /* |
1390 | * Steal bios from a request and add them to a bio list. | |
1391 | * The request must not have been partially completed before. | |
1392 | */ | |
1393 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
1394 | { | |
1395 | if (rq->bio) { | |
1396 | if (list->tail) | |
1397 | list->tail->bi_next = rq->bio; | |
1398 | else | |
1399 | list->head = rq->bio; | |
1400 | list->tail = rq->biotail; | |
1401 | ||
1402 | rq->bio = NULL; | |
1403 | rq->biotail = NULL; | |
1404 | } | |
1405 | ||
1406 | rq->__data_len = 0; | |
1407 | } | |
1408 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
1409 | ||
3bcddeac | 1410 | /** |
2e60e022 | 1411 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 1412 | * @req: the request being processed |
2a842aca | 1413 | * @error: block status code |
8ebf9756 | 1414 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
1415 | * |
1416 | * Description: | |
8ebf9756 RD |
1417 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
1418 | * the request structure even if @req doesn't have leftover. | |
1419 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
1420 | * |
1421 | * This special helper function is only for request stacking drivers | |
1422 | * (e.g. request-based dm) so that they can handle partial completion. | |
3a211b71 | 1423 | * Actual device drivers should use blk_mq_end_request instead. |
2e60e022 TH |
1424 | * |
1425 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
1426 | * %false return from this function. | |
3bcddeac | 1427 | * |
1954e9a9 BVA |
1428 | * Note: |
1429 | * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both | |
1430 | * blk_rq_bytes() and in blk_update_request(). | |
1431 | * | |
3bcddeac | 1432 | * Return: |
2e60e022 TH |
1433 | * %false - this request doesn't have any more data |
1434 | * %true - this request has more data | |
3bcddeac | 1435 | **/ |
2a842aca CH |
1436 | bool blk_update_request(struct request *req, blk_status_t error, |
1437 | unsigned int nr_bytes) | |
1da177e4 | 1438 | { |
f79ea416 | 1439 | int total_bytes; |
1da177e4 | 1440 | |
2a842aca | 1441 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); |
4a0efdc9 | 1442 | |
2e60e022 TH |
1443 | if (!req->bio) |
1444 | return false; | |
1445 | ||
54d4e6ab MG |
1446 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
1447 | if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ && | |
1448 | error == BLK_STS_OK) | |
1449 | req->q->integrity.profile->complete_fn(req, nr_bytes); | |
1450 | #endif | |
1451 | ||
2a842aca CH |
1452 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
1453 | !(req->rq_flags & RQF_QUIET))) | |
178cc590 | 1454 | print_req_error(req, error, __func__); |
1da177e4 | 1455 | |
bc58ba94 | 1456 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 1457 | |
f79ea416 KO |
1458 | total_bytes = 0; |
1459 | while (req->bio) { | |
1460 | struct bio *bio = req->bio; | |
4f024f37 | 1461 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 1462 | |
9c24c10a | 1463 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 1464 | req->bio = bio->bi_next; |
1da177e4 | 1465 | |
fbbaf700 N |
1466 | /* Completion has already been traced */ |
1467 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
f79ea416 | 1468 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 1469 | |
f79ea416 KO |
1470 | total_bytes += bio_bytes; |
1471 | nr_bytes -= bio_bytes; | |
1da177e4 | 1472 | |
f79ea416 KO |
1473 | if (!nr_bytes) |
1474 | break; | |
1da177e4 LT |
1475 | } |
1476 | ||
1477 | /* | |
1478 | * completely done | |
1479 | */ | |
2e60e022 TH |
1480 | if (!req->bio) { |
1481 | /* | |
1482 | * Reset counters so that the request stacking driver | |
1483 | * can find how many bytes remain in the request | |
1484 | * later. | |
1485 | */ | |
a2dec7b3 | 1486 | req->__data_len = 0; |
2e60e022 TH |
1487 | return false; |
1488 | } | |
1da177e4 | 1489 | |
a2dec7b3 | 1490 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
1491 | |
1492 | /* update sector only for requests with clear definition of sector */ | |
57292b58 | 1493 | if (!blk_rq_is_passthrough(req)) |
a2dec7b3 | 1494 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 1495 | |
80a761fd | 1496 | /* mixed attributes always follow the first bio */ |
e8064021 | 1497 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 1498 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 1499 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
1500 | } |
1501 | ||
ed6565e7 CH |
1502 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
1503 | /* | |
1504 | * If total number of sectors is less than the first segment | |
1505 | * size, something has gone terribly wrong. | |
1506 | */ | |
1507 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
1508 | blk_dump_rq_flags(req, "request botched"); | |
1509 | req->__data_len = blk_rq_cur_bytes(req); | |
1510 | } | |
2e46e8b2 | 1511 | |
ed6565e7 | 1512 | /* recalculate the number of segments */ |
e9cd19c0 | 1513 | req->nr_phys_segments = blk_recalc_rq_segments(req); |
ed6565e7 | 1514 | } |
2e46e8b2 | 1515 | |
2e60e022 | 1516 | return true; |
1da177e4 | 1517 | } |
2e60e022 | 1518 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 1519 | |
2d4dc890 IL |
1520 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1521 | /** | |
1522 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
1523 | * @rq: the request to be flushed | |
1524 | * | |
1525 | * Description: | |
1526 | * Flush all pages in @rq. | |
1527 | */ | |
1528 | void rq_flush_dcache_pages(struct request *rq) | |
1529 | { | |
1530 | struct req_iterator iter; | |
7988613b | 1531 | struct bio_vec bvec; |
2d4dc890 IL |
1532 | |
1533 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 1534 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
1535 | } |
1536 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
1537 | #endif | |
1538 | ||
ef9e3fac KU |
1539 | /** |
1540 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
1541 | * @q : the queue of the device being checked | |
1542 | * | |
1543 | * Description: | |
1544 | * Check if underlying low-level drivers of a device are busy. | |
1545 | * If the drivers want to export their busy state, they must set own | |
1546 | * exporting function using blk_queue_lld_busy() first. | |
1547 | * | |
1548 | * Basically, this function is used only by request stacking drivers | |
1549 | * to stop dispatching requests to underlying devices when underlying | |
1550 | * devices are busy. This behavior helps more I/O merging on the queue | |
1551 | * of the request stacking driver and prevents I/O throughput regression | |
1552 | * on burst I/O load. | |
1553 | * | |
1554 | * Return: | |
1555 | * 0 - Not busy (The request stacking driver should dispatch request) | |
1556 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
1557 | */ | |
1558 | int blk_lld_busy(struct request_queue *q) | |
1559 | { | |
344e9ffc | 1560 | if (queue_is_mq(q) && q->mq_ops->busy) |
9ba20527 | 1561 | return q->mq_ops->busy(q); |
ef9e3fac KU |
1562 | |
1563 | return 0; | |
1564 | } | |
1565 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
1566 | ||
78d8e58a MS |
1567 | /** |
1568 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
1569 | * @rq: the clone request to be cleaned up | |
1570 | * | |
1571 | * Description: | |
1572 | * Free all bios in @rq for a cloned request. | |
1573 | */ | |
1574 | void blk_rq_unprep_clone(struct request *rq) | |
1575 | { | |
1576 | struct bio *bio; | |
1577 | ||
1578 | while ((bio = rq->bio) != NULL) { | |
1579 | rq->bio = bio->bi_next; | |
1580 | ||
1581 | bio_put(bio); | |
1582 | } | |
1583 | } | |
1584 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
1585 | ||
1586 | /* | |
1587 | * Copy attributes of the original request to the clone request. | |
1588 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
1589 | */ | |
1590 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
b0fd271d | 1591 | { |
b0fd271d KU |
1592 | dst->__sector = blk_rq_pos(src); |
1593 | dst->__data_len = blk_rq_bytes(src); | |
297ba57d BVA |
1594 | if (src->rq_flags & RQF_SPECIAL_PAYLOAD) { |
1595 | dst->rq_flags |= RQF_SPECIAL_PAYLOAD; | |
1596 | dst->special_vec = src->special_vec; | |
1597 | } | |
b0fd271d KU |
1598 | dst->nr_phys_segments = src->nr_phys_segments; |
1599 | dst->ioprio = src->ioprio; | |
1600 | dst->extra_len = src->extra_len; | |
78d8e58a MS |
1601 | } |
1602 | ||
1603 | /** | |
1604 | * blk_rq_prep_clone - Helper function to setup clone request | |
1605 | * @rq: the request to be setup | |
1606 | * @rq_src: original request to be cloned | |
1607 | * @bs: bio_set that bios for clone are allocated from | |
1608 | * @gfp_mask: memory allocation mask for bio | |
1609 | * @bio_ctr: setup function to be called for each clone bio. | |
1610 | * Returns %0 for success, non %0 for failure. | |
1611 | * @data: private data to be passed to @bio_ctr | |
1612 | * | |
1613 | * Description: | |
1614 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
1615 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
1616 | * are not copied, and copying such parts is the caller's responsibility. | |
1617 | * Also, pages which the original bios are pointing to are not copied | |
1618 | * and the cloned bios just point same pages. | |
1619 | * So cloned bios must be completed before original bios, which means | |
1620 | * the caller must complete @rq before @rq_src. | |
1621 | */ | |
1622 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
1623 | struct bio_set *bs, gfp_t gfp_mask, | |
1624 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
1625 | void *data) | |
1626 | { | |
1627 | struct bio *bio, *bio_src; | |
1628 | ||
1629 | if (!bs) | |
f4f8154a | 1630 | bs = &fs_bio_set; |
78d8e58a MS |
1631 | |
1632 | __rq_for_each_bio(bio_src, rq_src) { | |
1633 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
1634 | if (!bio) | |
1635 | goto free_and_out; | |
1636 | ||
1637 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
1638 | goto free_and_out; | |
1639 | ||
1640 | if (rq->bio) { | |
1641 | rq->biotail->bi_next = bio; | |
1642 | rq->biotail = bio; | |
1643 | } else | |
1644 | rq->bio = rq->biotail = bio; | |
1645 | } | |
1646 | ||
1647 | __blk_rq_prep_clone(rq, rq_src); | |
1648 | ||
1649 | return 0; | |
1650 | ||
1651 | free_and_out: | |
1652 | if (bio) | |
1653 | bio_put(bio); | |
1654 | blk_rq_unprep_clone(rq); | |
1655 | ||
1656 | return -ENOMEM; | |
b0fd271d KU |
1657 | } |
1658 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
1659 | ||
59c3d45e | 1660 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
1661 | { |
1662 | return queue_work(kblockd_workqueue, work); | |
1663 | } | |
1da177e4 LT |
1664 | EXPORT_SYMBOL(kblockd_schedule_work); |
1665 | ||
ee63cfa7 JA |
1666 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) |
1667 | { | |
1668 | return queue_work_on(cpu, kblockd_workqueue, work); | |
1669 | } | |
1670 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
1671 | ||
818cd1cb JA |
1672 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
1673 | unsigned long delay) | |
1674 | { | |
1675 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
1676 | } | |
1677 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
1678 | ||
75df7136 SJ |
1679 | /** |
1680 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
1681 | * @plug: The &struct blk_plug that needs to be initialized | |
1682 | * | |
1683 | * Description: | |
40405851 JM |
1684 | * blk_start_plug() indicates to the block layer an intent by the caller |
1685 | * to submit multiple I/O requests in a batch. The block layer may use | |
1686 | * this hint to defer submitting I/Os from the caller until blk_finish_plug() | |
1687 | * is called. However, the block layer may choose to submit requests | |
1688 | * before a call to blk_finish_plug() if the number of queued I/Os | |
1689 | * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than | |
1690 | * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if | |
1691 | * the task schedules (see below). | |
1692 | * | |
75df7136 SJ |
1693 | * Tracking blk_plug inside the task_struct will help with auto-flushing the |
1694 | * pending I/O should the task end up blocking between blk_start_plug() and | |
1695 | * blk_finish_plug(). This is important from a performance perspective, but | |
1696 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
1697 | * for a memory allocation, memory reclaim could end up wanting to free a | |
1698 | * page belonging to that request that is currently residing in our private | |
1699 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
1700 | * this kind of deadlock. | |
1701 | */ | |
73c10101 JA |
1702 | void blk_start_plug(struct blk_plug *plug) |
1703 | { | |
1704 | struct task_struct *tsk = current; | |
1705 | ||
dd6cf3e1 SL |
1706 | /* |
1707 | * If this is a nested plug, don't actually assign it. | |
1708 | */ | |
1709 | if (tsk->plug) | |
1710 | return; | |
1711 | ||
320ae51f | 1712 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 1713 | INIT_LIST_HEAD(&plug->cb_list); |
5f0ed774 | 1714 | plug->rq_count = 0; |
ce5b009c | 1715 | plug->multiple_queues = false; |
5f0ed774 | 1716 | |
73c10101 | 1717 | /* |
dd6cf3e1 SL |
1718 | * Store ordering should not be needed here, since a potential |
1719 | * preempt will imply a full memory barrier | |
73c10101 | 1720 | */ |
dd6cf3e1 | 1721 | tsk->plug = plug; |
73c10101 JA |
1722 | } |
1723 | EXPORT_SYMBOL(blk_start_plug); | |
1724 | ||
74018dc3 | 1725 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
1726 | { |
1727 | LIST_HEAD(callbacks); | |
1728 | ||
2a7d5559 SL |
1729 | while (!list_empty(&plug->cb_list)) { |
1730 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 1731 | |
2a7d5559 SL |
1732 | while (!list_empty(&callbacks)) { |
1733 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
1734 | struct blk_plug_cb, |
1735 | list); | |
2a7d5559 | 1736 | list_del(&cb->list); |
74018dc3 | 1737 | cb->callback(cb, from_schedule); |
2a7d5559 | 1738 | } |
048c9374 N |
1739 | } |
1740 | } | |
1741 | ||
9cbb1750 N |
1742 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
1743 | int size) | |
1744 | { | |
1745 | struct blk_plug *plug = current->plug; | |
1746 | struct blk_plug_cb *cb; | |
1747 | ||
1748 | if (!plug) | |
1749 | return NULL; | |
1750 | ||
1751 | list_for_each_entry(cb, &plug->cb_list, list) | |
1752 | if (cb->callback == unplug && cb->data == data) | |
1753 | return cb; | |
1754 | ||
1755 | /* Not currently on the callback list */ | |
1756 | BUG_ON(size < sizeof(*cb)); | |
1757 | cb = kzalloc(size, GFP_ATOMIC); | |
1758 | if (cb) { | |
1759 | cb->data = data; | |
1760 | cb->callback = unplug; | |
1761 | list_add(&cb->list, &plug->cb_list); | |
1762 | } | |
1763 | return cb; | |
1764 | } | |
1765 | EXPORT_SYMBOL(blk_check_plugged); | |
1766 | ||
49cac01e | 1767 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 | 1768 | { |
74018dc3 | 1769 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
1770 | |
1771 | if (!list_empty(&plug->mq_list)) | |
1772 | blk_mq_flush_plug_list(plug, from_schedule); | |
73c10101 | 1773 | } |
73c10101 | 1774 | |
40405851 JM |
1775 | /** |
1776 | * blk_finish_plug - mark the end of a batch of submitted I/O | |
1777 | * @plug: The &struct blk_plug passed to blk_start_plug() | |
1778 | * | |
1779 | * Description: | |
1780 | * Indicate that a batch of I/O submissions is complete. This function | |
1781 | * must be paired with an initial call to blk_start_plug(). The intent | |
1782 | * is to allow the block layer to optimize I/O submission. See the | |
1783 | * documentation for blk_start_plug() for more information. | |
1784 | */ | |
73c10101 JA |
1785 | void blk_finish_plug(struct blk_plug *plug) |
1786 | { | |
dd6cf3e1 SL |
1787 | if (plug != current->plug) |
1788 | return; | |
f6603783 | 1789 | blk_flush_plug_list(plug, false); |
73c10101 | 1790 | |
dd6cf3e1 | 1791 | current->plug = NULL; |
73c10101 | 1792 | } |
88b996cd | 1793 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 1794 | |
1da177e4 LT |
1795 | int __init blk_dev_init(void) |
1796 | { | |
ef295ecf CH |
1797 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
1798 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
0762b23d | 1799 | FIELD_SIZEOF(struct request, cmd_flags)); |
ef295ecf CH |
1800 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
1801 | FIELD_SIZEOF(struct bio, bi_opf)); | |
9eb55b03 | 1802 | |
89b90be2 TH |
1803 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
1804 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 1805 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
1806 | if (!kblockd_workqueue) |
1807 | panic("Failed to create kblockd\n"); | |
1808 | ||
c2789bd4 | 1809 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 1810 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 1811 | |
18fbda91 OS |
1812 | #ifdef CONFIG_DEBUG_FS |
1813 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
1814 | #endif | |
1815 | ||
d38ecf93 | 1816 | return 0; |
1da177e4 | 1817 | } |